Database Query

Natural-language query over your data: ATG can run SQL against your database or a managed one, with read-only access and controlled schema exposure.

3 min read

Business value

Many questions your teams ask are about structured data: sales figures, inventory, project status, or any other information stored in a database. Writing SQL or waiting for a report is slow—and for many users, SQL is simply out of reach (too technical). Asking in plain language is faster and accessible to everyone who needs the answer. Database Query lets users phrase questions naturally and get answers from your data, without writing a single query.

Another common pain point is combining information that lives in different places: several SQL databases or schemas, structured tables alongside unstructured knowledge (documents, wikis, chat), or operational data vs. narrative context. The orchestrator agent can use the Database Query tool together with other tools and knowledge sources (e.g. Tailored Tools, organization documents) so one conversation can cross those boundaries instead of forcing users to jump between systems.

ATG can generate and run SQL against your own database or a database hosted and managed for you (for example a dedicated PostgreSQL schema). The assistant exposes this as a tool to the orchestrator agent: when a question is about this data, the agent calls the tool, which turns the question into SQL, runs it, and returns an interpreted result.

How it can be deployed

A typical setup is:

  • A dedicated client database (e.g. a dedicated PostgreSQL schema – an isolated set of tables in the same database). This keeps your data separate and manageable.
  • A dedicated Database Query tool exposed to the orchestrator agent (AI). The agent sees this tool among the others it can use and decides when to call it.

The tool is driven by three configuration texts:

  1. Tool description – Shown to the orchestrator agent so it knows when and how to call the tool (e.g. what kinds of questions it can answer).
  2. SQL generation instructions – A description of the schema (tables, columns, business context) used by a dedicated SQL-generation AI to produce the SQL from the user’s natural-language question.
  3. Result interpretation – A short text that is sent back to the orchestrator together with the query result, so it can interpret the data correctly (column meanings, units, business context).

The tool allows read-only access: only SELECT queries are permitted. No data is modified.

End-to-end flow

Loading diagram…

Multiple dialogue turns are possible; the Database Query tool is one among others.

Book a demo to discuss connecting your database or setting up a managed one for this capability.